[uClibc] buildroot and other questions
Ralph Siemsen
ralphs at netwinder.org
Tue Mar 29 18:57:58 UTC 2005
Szekely Gyorgy wrote:
> Floating points.
> I'm using a PXA255 processor in a project, no FPU. I compiled the NWFPE
> in the kernel, but I've read somewhere that this is slow because the
> processor generates an exception at every FP instruction. If gcc
> compiles calls to the floating point functions directly is a bit faster.
NWFPE isn't quite that bad - successive FP operations will be done with
a single exception. But still, software-float is much better/faster.
When you tell gcc "-msoft-float" then it will make a function call for
all float operations instead of putting in FP opcodes. You then have to
provide a libfloat where these functions are defined. Fortunately, in
gcc-3.4 this is already included.
With buildroot you need only select "Use software floating point by
default" (no need for libfloat as its inside gcc now). You can then
take our NWFPE from your kernel config if you like, as nothing should be
using it.
> What's multilib? Do I need it? (same project)
No. Its for mixed 32/64 bit library support.
> What's armeb?
That's big-endian ARM.
-Ralph
More information about the uClibc
mailing list